spaCy/spacy/pipeline/entity_linker.py
Ines Montani 43b960c01b
Refactor pipeline components, config and language data (#5759)
* Update with WIP

* Update with WIP

* Update with pipeline serialization

* Update types and pipe factories

* Add deep merge, tidy up and add tests

* Fix pipe creation from config

* Don't validate default configs on load

* Update spacy/language.py

Co-authored-by: Ines Montani <ines@ines.io>

* Adjust factory/component meta error

* Clean up factory args and remove defaults

* Add test for failing empty dict defaults

* Update pipeline handling and methods

* provide KB as registry function instead of as object

* small change in test to make functionality more clear

* update example script for EL configuration

* Fix typo

* Simplify test

* Simplify test

* splitting pipes.pyx into separate files

* moving default configs to each component file

* fix batch_size type

* removing default values from component constructors where possible (TODO: test 4725)

* skip instead of xfail

* Add test for config -> nlp with multiple instances

* pipeline.pipes -> pipeline.pipe

* Tidy up, document, remove kwargs

* small cleanup/generalization for Tok2VecListener

* use DEFAULT_UPSTREAM field

* revert to avoid circular imports

* Fix tests

* Replace deprecated arg

* Make model dirs require config

* fix pickling of keyword-only arguments in constructor

* WIP: clean up and integrate full config

* Add helper to handle function args more reliably

Now also includes keyword-only args

* Fix config composition and serialization

* Improve config debugging and add visual diff

* Remove unused defaults and fix type

* Remove pipeline and factories from meta

* Update spacy/default_config.cfg

Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>

* Update spacy/default_config.cfg

* small UX edits

* avoid printing stack trace for debug CLI commands

* Add support for language-specific factories

* specify the section of the config which holds the model to debug

* WIP: add Language.from_config

* Update with language data refactor WIP

* Auto-format

* Add backwards-compat handling for Language.factories

* Update morphologizer.pyx

* Fix morphologizer

* Update and simplify lemmatizers

* Fix Japanese tests

* Port over tagger changes

* Fix Chinese and tests

* Update to latest Thinc

* WIP: xfail first Russian lemmatizer test

* Fix component-specific overrides

* fix nO for output layers in debug_model

* Fix default value

* Fix tests and don't pass objects in config

* Fix deep merging

* Fix lemma lookup data registry

Only load the lookups if an entry is available in the registry (and if spacy-lookups-data is installed)

* Add types

* Add Vocab.from_config

* Fix typo

* Fix tests

* Make config copying more elegant

* Fix pipe analysis

* Fix lemmatizers and is_base_form

* WIP: move language defaults to config

* Fix morphology type

* Fix vocab

* Remove comment

* Update to latest Thinc

* Add morph rules to config

* Tidy up

* Remove set_morphology option from tagger factory

* Hack use_gpu

* Move [pipeline] to top-level block and make [nlp.pipeline] list

Allows separating component blocks from component order – otherwise, ordering the config would mean a changed component order, which is bad. Also allows initial config to define more components and not use all of them

* Fix use_gpu and resume in CLI

* Auto-format

* Remove resume from config

* Fix formatting and error

* [pipeline] -> [components]

* Fix types

* Fix tagger test: requires set_morphology?

Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
Co-authored-by: svlandeg <sofie.vanlandeghem@gmail.com>
Co-authored-by: Matthew Honnibal <honnibal+gh@gmail.com>
2020-07-22 13:42:59 +02:00

367 lines
15 KiB
Python

from typing import Optional, Iterable, Callable, Dict, Iterator, Union, List, Tuple
from pathlib import Path
import srsly
import random
from thinc.api import CosineDistance, get_array_module, Model, Optimizer, Config
from thinc.api import set_dropout_rate
import warnings
from ..kb import KnowledgeBase
from ..tokens import Doc
from .pipe import Pipe, deserialize_config
from ..language import Language
from ..vocab import Vocab
from ..gold import Example
from ..errors import Errors, Warnings
from .. import util
default_model_config = """
[model]
@architectures = "spacy.EntityLinker.v1"
[model.tok2vec]
@architectures = "spacy.HashEmbedCNN.v1"
pretrained_vectors = null
width = 96
depth = 2
embed_size = 300
window_size = 1
maxout_pieces = 3
subword_features = true
dropout = null
"""
DEFAULT_NEL_MODEL = Config().from_str(default_model_config)["model"]
@Language.factory(
"entity_linker",
requires=["doc.ents", "doc.sents", "token.ent_iob", "token.ent_type"],
assigns=["token.ent_kb_id"],
default_config={
"kb": None, # TODO - what kind of default makes sense here?
"labels_discard": [],
"incl_prior": True,
"incl_context": True,
"model": DEFAULT_NEL_MODEL,
},
)
def make_entity_linker(
nlp: Language,
name: str,
model: Model,
kb: Optional[KnowledgeBase],
*,
labels_discard: Iterable[str],
incl_prior: bool,
incl_context: bool,
):
return EntityLinker(
nlp.vocab,
model,
name,
kb=kb,
labels_discard=labels_discard,
incl_prior=incl_prior,
incl_context=incl_context,
)
class EntityLinker(Pipe):
"""Pipeline component for named entity linking.
DOCS: https://spacy.io/api/entitylinker
"""
NIL = "NIL" # string used to refer to a non-existing link
def __init__(
self,
vocab: Vocab,
model: Model,
name: str = "entity_linker",
*,
kb: KnowledgeBase,
labels_discard: Iterable[str],
incl_prior: bool,
incl_context: bool,
) -> None:
self.vocab = vocab
self.model = model
self.name = name
cfg = {
"kb": kb,
"labels_discard": list(labels_discard),
"incl_prior": incl_prior,
"incl_context": incl_context,
}
self.kb = kb
if self.kb is None:
# create an empty KB that should be filled by calling from_disk
self.kb = KnowledgeBase(vocab=vocab)
else:
del cfg["kb"] # we don't want to duplicate its serialization
if not isinstance(self.kb, KnowledgeBase):
raise ValueError(Errors.E990.format(type=type(self.kb)))
self.cfg = dict(cfg)
self.distance = CosineDistance(normalize=False)
# how many neightbour sentences to take into account
self.n_sents = cfg.get("n_sents", 0)
def require_kb(self) -> None:
# Raise an error if the knowledge base is not initialized.
if len(self.kb) == 0:
raise ValueError(Errors.E139.format(name=self.name))
def begin_training(
self,
get_examples: Callable = lambda: [],
pipeline: Optional[List[Tuple[str, Callable[[Doc], Doc]]]] = None,
sgd: Optional[Optimizer] = None,
) -> Optimizer:
self.require_kb()
nO = self.kb.entity_vector_length
self.set_output(nO)
self.model.initialize()
if sgd is None:
sgd = self.create_optimizer()
return sgd
def update(
self,
examples: Iterable[Example],
*,
set_annotations: bool = False,
drop: float = 0.0,
sgd: Optional[Optimizer] = None,
losses: Optional[Dict[str, float]] = None,
) -> Dict[str, float]:
self.require_kb()
if losses is None:
losses = {}
losses.setdefault(self.name, 0.0)
if not examples:
return losses
sentence_docs = []
try:
docs = [eg.predicted for eg in examples]
except AttributeError:
types = set([type(eg) for eg in examples])
raise TypeError(
Errors.E978.format(name="EntityLinker", method="update", types=types)
)
if set_annotations:
# This seems simpler than other ways to get that exact output -- but
# it does run the model twice :(
predictions = self.model.predict(docs)
for eg in examples:
sentences = [s for s in eg.predicted.sents]
kb_ids = eg.get_aligned("ENT_KB_ID", as_string=True)
for ent in eg.predicted.ents:
kb_id = kb_ids[
ent.start
] # KB ID of the first token is the same as the whole span
if kb_id:
try:
# find the sentence in the list of sentences.
sent_index = sentences.index(ent.sent)
except AttributeError:
# Catch the exception when ent.sent is None and provide a user-friendly warning
raise RuntimeError(Errors.E030)
# get n previous sentences, if there are any
start_sentence = max(0, sent_index - self.n_sents)
# get n posterior sentences, or as many < n as there are
end_sentence = min(len(sentences) - 1, sent_index + self.n_sents)
# get token positions
start_token = sentences[start_sentence].start
end_token = sentences[end_sentence].end
# append that span as a doc to training
sent_doc = eg.predicted[start_token:end_token].as_doc()
sentence_docs.append(sent_doc)
set_dropout_rate(self.model, drop)
if not sentence_docs:
warnings.warn(Warnings.W093.format(name="Entity Linker"))
return 0.0
sentence_encodings, bp_context = self.model.begin_update(sentence_docs)
loss, d_scores = self.get_similarity_loss(
sentence_encodings=sentence_encodings, examples=examples
)
bp_context(d_scores)
if sgd is not None:
self.model.finish_update(sgd)
losses[self.name] += loss
if set_annotations:
self.set_annotations(docs, predictions)
return losses
def get_similarity_loss(self, examples: Iterable[Example], sentence_encodings):
entity_encodings = []
for eg in examples:
kb_ids = eg.get_aligned("ENT_KB_ID", as_string=True)
for ent in eg.predicted.ents:
kb_id = kb_ids[ent.start]
if kb_id:
entity_encoding = self.kb.get_vector(kb_id)
entity_encodings.append(entity_encoding)
entity_encodings = self.model.ops.asarray(entity_encodings, dtype="float32")
if sentence_encodings.shape != entity_encodings.shape:
err = Errors.E147.format(
method="get_similarity_loss", msg="gold entities do not match up"
)
raise RuntimeError(err)
gradients = self.distance.get_grad(sentence_encodings, entity_encodings)
loss = self.distance.get_loss(sentence_encodings, entity_encodings)
loss = loss / len(entity_encodings)
return loss, gradients
def __call__(self, doc: Doc) -> Doc:
kb_ids = self.predict([doc])
self.set_annotations([doc], kb_ids)
return doc
def pipe(self, stream: Iterable[Doc], batch_size: int = 128) -> Iterator[Doc]:
for docs in util.minibatch(stream, size=batch_size):
kb_ids = self.predict(docs)
self.set_annotations(docs, kb_ids)
yield from docs
def predict(self, docs):
""" Return the KB IDs for each entity in each doc, including NIL if there is no prediction """
self.require_kb()
entity_count = 0
final_kb_ids = []
if not docs:
return final_kb_ids
if isinstance(docs, Doc):
docs = [docs]
for i, doc in enumerate(docs):
sentences = [s for s in doc.sents]
if len(doc) > 0:
# Looping through each sentence and each entity
# This may go wrong if there are entities across sentences - which shouldn't happen normally.
for sent_index, sent in enumerate(sentences):
if sent.ents:
# get n_neightbour sentences, clipped to the length of the document
start_sentence = max(0, sent_index - self.n_sents)
end_sentence = min(
len(sentences) - 1, sent_index + self.n_sents
)
start_token = sentences[start_sentence].start
end_token = sentences[end_sentence].end
sent_doc = doc[start_token:end_token].as_doc()
# currently, the context is the same for each entity in a sentence (should be refined)
sentence_encoding = self.model.predict([sent_doc])[0]
xp = get_array_module(sentence_encoding)
sentence_encoding_t = sentence_encoding.T
sentence_norm = xp.linalg.norm(sentence_encoding_t)
for ent in sent.ents:
entity_count += 1
to_discard = self.cfg.get("labels_discard", [])
if to_discard and ent.label_ in to_discard:
# ignoring this entity - setting to NIL
final_kb_ids.append(self.NIL)
else:
candidates = self.kb.get_candidates(ent.text)
if not candidates:
# no prediction possible for this entity - setting to NIL
final_kb_ids.append(self.NIL)
elif len(candidates) == 1:
# shortcut for efficiency reasons: take the 1 candidate
# TODO: thresholding
final_kb_ids.append(candidates[0].entity_)
else:
random.shuffle(candidates)
# this will set all prior probabilities to 0 if they should be excluded from the model
prior_probs = xp.asarray(
[c.prior_prob for c in candidates]
)
if not self.cfg.get("incl_prior"):
prior_probs = xp.asarray(
[0.0 for c in candidates]
)
scores = prior_probs
# add in similarity from the context
if self.cfg.get("incl_context"):
entity_encodings = xp.asarray(
[c.entity_vector for c in candidates]
)
entity_norm = xp.linalg.norm(
entity_encodings, axis=1
)
if len(entity_encodings) != len(prior_probs):
raise RuntimeError(
Errors.E147.format(
method="predict",
msg="vectors not of equal length",
)
)
# cosine similarity
sims = xp.dot(
entity_encodings, sentence_encoding_t
) / (sentence_norm * entity_norm)
if sims.shape != prior_probs.shape:
raise ValueError(Errors.E161)
scores = (
prior_probs + sims - (prior_probs * sims)
)
# TODO: thresholding
best_index = scores.argmax().item()
best_candidate = candidates[best_index]
final_kb_ids.append(best_candidate.entity_)
if not (len(final_kb_ids) == entity_count):
err = Errors.E147.format(
method="predict", msg="result variables not of equal length"
)
raise RuntimeError(err)
return final_kb_ids
def set_annotations(self, docs: Iterable[Doc], kb_ids: List[int]) -> None:
count_ents = len([ent for doc in docs for ent in doc.ents])
if count_ents != len(kb_ids):
raise ValueError(Errors.E148.format(ents=count_ents, ids=len(kb_ids)))
i = 0
for doc in docs:
for ent in doc.ents:
kb_id = kb_ids[i]
i += 1
for token in ent:
token.ent_kb_id_ = kb_id
def to_disk(self, path: Union[str, Path], exclude: Iterable[str] = tuple()) -> None:
serialize = {}
self.cfg["entity_width"] = self.kb.entity_vector_length
serialize["cfg"] = lambda p: srsly.write_json(p, self.cfg)
serialize["vocab"] = lambda p: self.vocab.to_disk(p)
serialize["kb"] = lambda p: self.kb.dump(p)
serialize["model"] = lambda p: self.model.to_disk(p)
util.to_disk(path, serialize, exclude)
def from_disk(
self, path: Union[str, Path], exclude: Iterable[str] = tuple()
) -> "EntityLinker":
def load_model(p):
try:
self.model.from_bytes(p.open("rb").read())
except AttributeError:
raise ValueError(Errors.E149)
def load_kb(p):
self.kb = KnowledgeBase(
vocab=self.vocab, entity_vector_length=self.cfg["entity_width"]
)
self.kb.load_bulk(p)
deserialize = {}
deserialize["vocab"] = lambda p: self.vocab.from_disk(p)
deserialize["cfg"] = lambda p: self.cfg.update(deserialize_config(p))
deserialize["kb"] = load_kb
deserialize["model"] = load_model
util.from_disk(path, deserialize, exclude)
return self
def rehearse(self, examples, sgd=None, losses=None, **config):
raise NotImplementedError
def add_label(self, label):
raise NotImplementedError